Ref: https://learn.cantrill.io/courses/1820301/lectures/42177865

• Summary:
• Contents:

Amazon Kendra - Key Concepts

• 🔧 Intelligent text search service
  • Designed to mimic interacting with a human expert
  • Extracts answers from within a document
• Supports wide range of question types
  • Factoid: who, what, where
  • Descriptive
    • e.g. “how do I get my cat to stop being a jerk?” xD
  • Keyword
    • e.g. “what time is the keynote address?” → “address” can have multiple meanings… Kendra helps determine intent.
• Key components
  • (Knowledge) index: searchable block of data organized in an efficient way
    • Powered by ML
    • Kendra searches for indexes when dealing with queries
  • Data source: where your data lives
    • Kendra connects and indexes from this location
    • e.g. AWS internal (S3, RDS, WorkDocs, FSx…) or external (Confluence, Google Workspace, OneDrive, Salesforce… Kendra Web Crawler can even discover data from websites)
    • List of all possible data sources: https://docs.aws.amazon.com/kendra/latest/dg/hiw-data-source.html
  • Kendra synchronizes data sources with indexes based on a schedule (keeps index current)
  • Data sources contain documents, which are indexed
    • Structured documents (FAQs)
    • Unstructured documents (e.g. HTML, PDFs, text…)
    • List of all possible document types: https://docs.aws.amazon.com/kendra/latest/dg/index-document-types.html
• Other features
  • Incremental Learning: can learn from user interactions/feedback to promoted preferred results
  • Manual fine-tuning of search results (importance of data, freshness…)